djot-0.1.4.3: CHANGELOG.md

# Revision history for djot

## 0.1.4.3 -- 2026-09-23

  * Fix dropped attribute when attaching to last word of merged Strs.
    E.g. in `x y.z{.c}`, the class was silently discarded.

  * Fix implicit reference labels spanning multiple lines.
    Two bugs conspired to make `[link\ntext][]` resolve to an empty href
    even when a "link text" reference was defined.

  * Fix resumable attribute parser ignoring new input on resume.
    Previously a multiline block attribute followed by an indented
    block never finished parsing and the attributes were silently
    dropped, e.g.:

        {#id .class
          style="color:red"}
          A paragraph

  * Avoid quadratic inline parsing by caching chunk count. Previously
    parsing a paragraph was quadratic in the number of lines.

  * Speed up HTML escaping by copying unescaped runs wholesale.

  * Wrap letter list markers around after 26.

  * Fix off-by-one in tab column computation.

  * Propagate inline parse errors in `parseTextLines` instead of crashing.

  * Make NoPos an identity for `<>`, so Monoid Pos is lawful.
    Previously NoPos was absorbing `(NoPos <> p = NoPos)`,
    which would silently discard positions in any law-relying generic code
    (mconcat, fold). The only internal uses of `<>` on Pos (merging
    adjacent Strs) always combine two Pos or two NoPos values, so this
    does not change any parsing or rendering behavior.

  * Fix UTF-8 truncation in `inlinesToByteString`.

  * Improve classification of autolinks. Previously `<user.name@example.com>`
    was rendered as a plain URL link, not an email link. It's an email
    link if it contains an '@' preceded by a character other than ':'.

  * Preserve soft breaks for CR-only line endings.

  * Add tests for whitespace collapsing in quoted attribute values.

  * Fix off-by-one in `pAtMost`.

  * Skip empty chunks when advancing to the next chunk.

  * Speed up djot rendering of Str inlines:

    + Rewrite `escapeDjot` to work directly on ByteStrings, copying
      runs of unescapable characters wholesale instead of processing
      character by character through String. (Equivalence with the
      old implementation checked with 100,000 QuickCheck cases.)
    + Skip the smart-punctuation replacement passes when the
      ByteString contains no 0xE2 byte (the first byte of all the
      UTF-8 sequences involved), avoiding five `T.replace` traversals
      per Str in the common case.
    + Avoid emitting an empty literal after each single-space chunk.

## 0.1.4.2 -- 2026-08-27

  * Djot renderer: emit raw blocks/inlines for formats other than djot (#18).

  * Djot renderer: Fix calculation of fence length. We needed to add
    one to the length of the biggest backtick string in the code.

  * Fix two bugs in footnote handling:

    - Allow spaces and newlines in footnote references, as with djot.js.
    - Don't put a duplicate id on a second reference to same note.

## 0.1.4.1 -- 2026-07-21

  * Fix bug in verbatim parsing (#17). We were wrongly allowing
    backslash escapes for backticks.

  * Fix resolution of reference links with newlines (#16).

## 0.1.4 -- 2026-03-17

  * Ensure that delims aren't matched in link destinations (#15).

  * Change behavior for empty link label. To match djot.js; we now
    take the string content (rather than raw source) of the link
    description, so e.g. emphasis is ignored.

  * Html renderer: don't emit empty href for undefined ref link.
    This conforms to djot.js behavior, expected in tests.

  * Update links_and_images test from reference impl: djot.js.

## 0.1.3 -- 2026-02-01

* Fix nested sections bug when sections contain only lists (#14).
  When a section contained only list elements (no paragraphs),
  subsequent headings at the same level were incorrectly nested
  inside the previous section instead of being siblings.

* CLI: add `--version` option (#13).

* Export version from Djot module. [API change]

* Remove some INLINE pragmas. These don't make a measurable
  difference in benchmarks.

* There can't be a blank line btw block attrs and block.
  Cf. https://github.com/jgm/djot.js/issues/118

## 0.1.2.4 -- 2025-11-30

* Ensure that `'95--'96` doesn't get parsed as singlequoted.

* Properly handle bare `'}` for right single-quote (#12).

## 0.1.2.3 -- 2025-09-27

* Fix swallowing of indentation in code under blockquote (#11).

## 0.1.2.2 -- 2024-10-04

* Allow list items with blank lines between divs (#10).

* Fix parsing of indented tables (#8).

## 0.1.2.1 -- 2024-06-24

* Djot writer: include separator line in table when the table has
  non-default alignments but no header (#7).

## 0.1.2 -- 2024-05-10

* Allow `_` in symbols (see jgm/djot#296).

* Add Lift derivations to AST datatypes (#5, Gideon Farrell) [API change].

## 0.1.1.3 -- 2024-03-17

* Ensure that tables end when we hit a blank line (#4).

* Fix parsing of table immediately after list (#4).

## 0.1.1.2 -- 2024-03-14

* Fix bug parsing regular paragraphs after list (#4).

## 0.1.1.1 -- 2024-03-03

* Revert "Djot.Blocks: use ByteString directly in `toIdentifier` (#1)"
  This caused problems for UTF-8 sequences that contained the
  byte 0xa0, which B8.words treats as a space character.

* AST: avoid using B8.words in normalizeLabel.

* Avoid using isSpace in attribute parsing. isSpace matches a byte 0x0a,
  which can break up a UTF-8 sequence. Limit to ASCII whitespace.

  * Add test with UTF-8 identifier. See jgm/pandoc#9541.

## 0.1.1.0 -- 2024-02-29

* Add Data instances to everything in the AST [API change].

* Ensure that block attributes are indented on subsequent lines (#2).

* Djot.Blocks: use ByteString directly in `toIdentifier` (#1,
  Vaibhav Sagar).

## 0.1.0.0 -- 2024-02-14

* Initial release.